{
  "markdown": "# Probabl Skills\n\nA set of skills to steer your AI-assisted machine learning experiments.\nThe skills help you:\n\n- build your machine learning pipeline with core data science libraries\n  (e.g. scikit-learn, skrub, skore, pandas, polar) while ensuring\n  your agent follows correct methodologies\n- evaluate and store your results so you can easily audit and get insights from them\n- connect your agent to [Skore Hub](https://skore.probabl.ai/) to get a comprehensive view of\n  your experiments and results\n- iterate on your next experiments using insights from Skore diagnostics and your own\n  feedback\n- organize your workspace according to best practices for data science projects\n  (e.g. cookiecutter template)\n\nProbabl skills let you focus on the science while AI agents handle the implementation,\nguided by two important ingredients: core data science libraries for maintainability  and\nmethodological best practices for running your machine learning experiments properly.\n\nIn practice, from a prompt such as:\n\n```text\n╭────────────────────────────────────────────────────────────────────────╮\n│ > Given the context in the file `data/README.md` and the data located  │\n│   in `data/`, let's build a first machine learning pipeline that will  │\n│   serve as baseline for the next experiments that we are going to run  │\n│   together.                                                            │\n╰────────────────────────────────────────────────────────────────────────╯\n```\n\nyou can expect your agent to start experimenting with you. The skills work well with\nmodels such as Claude Opus and Sonnet and produce great results with smaller models such\nas Qwen 3.6 30B or DeepSeek v4 Flash.\n\nAs for agent harnesses, we tested them with Claude Code, OpenCode, Cursor, and GitHub\nCopilot and found no significant difference in terms of skill invocation.\n\n## Install\n\nYou can install the skills using the `skore` CLI that you can install from PyPI or from\nconda-forge and run the following command.\n\nFirst install [skore-cli](https://github.com/probabl-ai/skore-cli):\n```\n# with pip\npip install skore-cli\n# with uv\nuv tool install skore-cli\n# with pixi\npixi global install skore-cli\n```\n\nThen run the following command:\n\n```bash\nskore skills install\n```\n\nYou can use `uvx` or `pixi exec` to install the `skore` CLI and directly run the\ncommand in an isolated environment:\n\n```bash\nuvx --from skore-cli skore skills install\n```\n\nor\n\n```bash\npixi exec --spec skore-cli skore skills install\n```\n\nIf you prefer `npx`, then you can use:\n\n```bash\nnpx skills add probabl-ai/skills\n```\n\n### Alternative — Claude Code plugin marketplace\n\nIf you only use Claude Code and prefer the native plugin flow, this repo is\nalso a [Claude Code plugin marketplace](https://docs.claude.com/en/docs/claude-code/plugin-marketplaces):\n\n```bash\n/plugin marketplace add probabl-ai/skills\n```\n\n```bash\n/plugin install probabl-skills@probabl-skills\n```\n\n`/plugin update` pulls new releases.\n\n## Skills in detail\n\n### ML pipeline lifecycle\n\n| Skill | Description |\n| --- | --- |\n| [explore-ml-data](skills/explore-ml-data/SKILL.md) | Explore the dataset before designing any model. |\n| [build-ml-pipeline](skills/build-ml-pipeline/SKILL.md) | Build a machine learning pipeline from the data source to the learner, including multi-tables engineering. |\n| [evaluate-ml-pipeline](skills/evaluate-ml-pipeline/SKILL.md) | Evaluate a complex machine learning pipeline and get structured reports including metrics, plots, and diagnostics. |\n| [test-ml-pipeline](skills/test-ml-pipeline/SKILL.md) | Make sure that your machine learning pipeline is production-ready statistically and functionally. |\n| [smoke-test-ml-pipeline](skills/smoke-test-ml-pipeline/SKILL.md) | Stress test your machine learning pipeline on future data to make sure it works. |\n| [audit-ml-pipeline](skills/audit-ml-pipeline/SKILL.md) | Once testing and the experiment are done, audit the model by loading a skore report and investigate. |\n\n### Iteration loop\n\n| Skill | Description |\n| --- | --- |\n| [iterate-ml-experiment](skills/iterate-ml-experiment/SKILL.md) | Design, keep track of experiments, and iterate on them. |\n| [iterate-from-skore](skills/iterate-from-skore/SKILL.md) | Use skore to run diagnostics and checks that can be reported and addressed in the next experiment. |\n| [iterate-from-user](skills/iterate-from-user/SKILL.md) | As a user, be in the loop and propose new experiments — free-text, a scientific article URL, or a resource link (GitHub issue / spec / reference repo). |\n\n### Workspace and tooling\n\n| Skill | Description |\n| --- | --- |\n| [organize-ml-workspace](skills/organize-ml-workspace/SKILL.md) | An organized workspace to keep track of your experiments. |\n| [python-code-style](skills/python-code-style/SKILL.md) | Enforce best practices out-of-the-box from the Python ecosystem in your code. |\n| [python-env-manager](skills/python-env-manager/SKILL.md) | Bootstrapping the experiment setup based on your favorite Python environment manager. |\n| [data-science-python-stack](skills/data-science-python-stack/SKILL.md) | Opinionated one-library-per-job Python stack, organized into mandatory / user-choice / optional / transitive tiers. |\n\n### API references\n\n| Skill | Description |\n| --- | --- |\n| [python-api](skills/python-api/SKILL.md) | Discover the public API of any installed Python package to help agents navigate without polluting your workspace. |\n",
  "bytes": 5423,
  "sha": "31aaa9209aad90f484338ab9d36e067f9f15e06074661447ced33617f7dd073a",
  "repo_slug": "probabl-ai/skills",
  "fonte": "repo",
  "truncated": false,
  "api": "https://agentalog.com/api/listings/plg_probabl_ai_skills_probabl_skills_8fcea950/readme"
}